Toward a standardized quantitative and qualitative insect monitoring scheme

Abstract The number of insect species and insect abundances decreased severely during the past decades over major parts of Central Europe. Previous studies documented declines of species richness, abundances, shifts in species composition, and decreasing biomass of flying insects. In this study, we present a standardized approach to quantitatively and qualitatively assess insect diversity, biomass, and the abundance of taxa, in parallel. We applied two methods: Malaise traps, and automated and active light trapping. Sampling was conducted from April to October 2018 in southern Germany, at four sites representing conventional and organic farming. Bulk samples obtained from Malaise traps were further analyzed using DNA metabarcoding. Larger moths (Macroheterocera) collected with light trapping were further classified according to their degree of endangerment. Our methods provide valuable quantitative and qualitative data. Our results indicate more biomass and higher species richness, as well as twice the number of Red List lepidopterans in organic farmland than in conventional farmland. This combination of sampling methods with subsequent DNA metabarcoding and assignments of individuals according depending on ecological characteristics and the degree of endangerment allows to evaluate the status of landscapes and represents a suitable setup for large‐scale long‐term insect monitoring across Central Europe, and elsewhere.

trends have been detectable since the 1950s, while major losses occurred during the past two decades (Habel et al., 2016). Hereby, species with specific habitat requirements (e.g., species demanding specific food plants or habitat structures) decreased particularly during the past years (Sodhi, Brook, & Bradshaw, 2009). But, also local populations and thus densities and abundances of rather generalist species, using a large variety of resources, decreased significantly (Sodhi et al., 2009). This trend was exemplary shown by the 75% reduction of biomass from flying insects over past three decades (Hallmann et al., 2017).
Previous studies on insect decline mainly focused on one single proxy, for example species richness, species abundance, species composition, or biomass (Sanders & Hess, 2019), but rarely considered all parameters in parallel. Furthermore, most existing studies on insect decline show various shortcomings (Saunders, 2017). (a) Studies refer to single species only, and thus, the validity of general trends is questionable (Reichholf, 2005(Reichholf, , 2006(Reichholf, , 2008; (b) studies only consider few time steps (Augenstein, Ulrich, & Habel, 2012;Filz, Engler, Stoffels, Weitzel, & Schmitt, 2013;Hallmann et al., 2017;Wenzel et al., 2006) and thus obtained changes in species richness, and community composition might be not representative for a larger time period; (c) data collected mostly refer to a geographically restricted area (Habel et al., 2016) cannot be translated to other regions; and (d) most long-term data sets are of low quality, with multiple data gaps (Desender, Dekoninck, Dufrêne, & Maes, 2010). Most studies present more than one of these limitations (McGill, Dornelas, Gotelli, & Magurran, 2015). Thus, there is an urgent need to set a standardized insect monitoring scheme across Central Europe, applying consistent techniques and sampling protocols.
In this study, we combine quantitative and qualitative approaches to study trends of (a) insect diversity, (b) insect abundances, (c) insect community assembly, and (d) insect biomass. Insects were collected with Malaise traps, in combination with automated and active light trapping, at identical sites. Subsequently, we weighted the biomass and used DNA metabarcoding to calculate species richness. Catches of the light trapping were further analyzed in respect of species abundances and community composition, considering species-specific traits. With our data collected during this first year of monitoring, we will highlight the following questions: 1. Are the methods selected suitable to assess quantitative and qualitative parameters? 2. Do data obtained from the two approaches provide congruent or diverging trends?

| Study area
Our study area is located in southern Germany, 15 km distant to the city Pfaffenhofen. We established four study sites, two in organic, and two in conventional farmland. Malaise traps and light traps were installed at extensively used grasslands. Both grasslands border north-south forest fringes and were mowed twice a year (June 24 and October 07 in organic farmland; and June 01 and September 07 in conventional farmland) and without any application of pesticides, however with the use of organic fertilizers. The organic farmland was cultivated without any pesticides since 64 years, but with organic fertilizer (on February 05). Artificial fertilizers were used on the conventional farmland (in 2018 on February 14 and April 05). The following pesticides were applied on the bordering rye and cornfields in conventional farmland: Broadway (130 g 0.5 L/ha;14.4.2018), Gardo Gold (3 L/ha;27.5.2018), Callisto (0.75 L/ha;27.5.2018), and the shortcut Chlormequat (0.3 L/ha; 12.5.2018). Distance between organic and conventional grasslands was about 700 m.

| Biomass
Dry and wet biomass material was weighted and analyzed separately, according to Ssymanck et al. (2018). Biomasses of macrolepidopterans and orthopterans have been separated and weighted before, and were subsequently added to the total weight. Based on the standardized methodology of Sorg, Schwan, Stenmans, and Müller (2013), species were dried according to size selection using a sieve (6.5 mm) in diameter in a 70°C oven over night (or at least for 8 hr).

| Metabarcoding
Species identification of organic material in the Malaise traps was performed using DNA metabarcoding. However, macrolepidopterans and orthopterans were a priori sorted out of the Malaise traps and identified by experts at the ZSM. Some macrolepidopterans or fragments of them remained in the bulk sample and were subsequently detected by the metabarcoding approach. All microlepidopterans were identified with the metabarcoding approach in the bulk sample. Each single dried sample (altogether 4 × 11 = 44 samples) was homogenized in a FastPrep96 machine (MP Biomedicals) using sterile steal beads in order to generate a homogeneous mixture of arthropods and submit- The bioinformatics processing of raw FASTQ files from Illumina was carried out using the VSEARCH suite v2.9.1 (Rognes, Flouri, Nichols, Quince, & Mahé, 2016) and Cutadapt v1.18 (Martin, 2011). Forward and reverse reads in each sample were merged using the VSEARCH program "fastq_mergepairs" with a minimum overlap of 10 bp, yielding approximately 313 bp sequences. Forward and reverse primers were removed with Cutadapt, using the "discard_untrimmed" option to discard sequences for which primers were not reliably detected at ≥90% identity. Quality filtering was done with the "fastq_filter" in VSEARCH, keeping sequences with zero expected errors ("fastq_maxee" 1).
Sequences were dereplicated with "derep_fulllength," first at the sample level, and then concatenated into one FASTA file, which was subsequently dereplicated. Chimeric sequences were filtered out from the FASTA file using the VSEARCH program "uchime_denovo." The remaining sequences were then clustered into OTUs at 97% identity with "cluster_size," a greedy centroid-based clustering program. OTUs were blasted against a custom Animalia database downloaded from BOLD on 28 November 2018, including taxonomy and BIN information, by means of Geneious (v.10.2.5-Biomatters, Auckland-New Zealand), and following methods described in Morinière et al. (2016).
The resulting csv file which included the OTU ID, BOLD Process ID, BIN, Hit-%-ID value (percentage of overlap similarity (identical basepairs) of an OTU query sequence with its closest counterpart in the database), length of the top BLAST hit sequence, phylum, class, order, family, genus, and species information for each detected OTU was exported from Geneious and combined with the OTU table generated by the bioinformatic pipeline. The combined results table was then filtered by Hit-%-ID value and total read numbers per OTU. All entries with identifications below 97% and total read numbers below 0.01% of the summed reads per sample were removed from the analysis. OTUs were then assigned to the respective BIN. Additionally, the API provided by BOLD was used to retrieve BIN species and BIN countries for every OTU, and the Hit-%-IDs were aggregated over OTUs that found a hit in the same BIN and shown in the corresponding column as % range (Table S2). To validate the BOLD BLAST results, a separate BLAST search was carried out in Geneious (using the same parameters) against a local copy of the NCBI nucleotide database downloaded from ftp://ftp.ncbi.nlm.nih.gov/blast /db/ (see Table "BIN sharing and countries" in Table S2) on 28 November 2018. Interactive Krona charts were produced from the taxonomic information using KronaTools v1.3 (Ondov, Bergman, & Phillippy, 2011) Species identification in the Malaise trap samples was based on high-throughput sequencing (HTS) data grouped to genetic clusters (OTUs), blasted, and assigned to barcode index numbers ("BINs": Ratnasingham & Hebert, 2013) which are considered to be a good proxy for species numbers (Hausmann et al., 2013;Ratnasingham & Hebert, 2013). In our case, the detailed analysis of the Lepidoptera data revealed that the frequency of "false positives" (0.5%) and BINsharing (1.5%) obstructing species discrimination (but nevertheless still pointing to species complexes) played a negligible role (see results for details).

| Species composition and phenologies
We used two approaches to infer differences in species composition between organic and conventional farming sites. First, we used a random sampling model and calculated for all taxa having at least 15 OTUs the probability of having k joint OTUs, given that the organic farming sites had l and the conventional farming sites m OTUs, while the total number (the local pool size) was assumed as n = l + m − k.
This probability is given by (Connor & Simberloff, 1978): and has the random expectation of k exp = lm/n OTUs. Significant differences ∆k = k exp − k point to differences in community composition.
We note that the observed probabilities p strongly depend on the pool size n and cannot be compared among taxa directly. Therefore, we also estimated from Equation (1) the required number of OTUs, n req necessary to obtain the observed k at the 5% error level. From n req , we also obtained the (minimal) degree of undersampling = 100 (1 − n/n req ).
We calculated Spearman's rank-order correlations (r S ) between all species, which jointly occur at both farming sites. Significantly, negative r S values indicate structural differences between the two communities in terms of relative abundances. We used one-way ANOVA to infer the difference in extinction probabilities between both farming types using average OTU abundance at both sites as the dependent variable.
To infer whether organic and conventional farming influence the phenology of arthropods, we first analyzed the combined phenologies of six major arthropod taxa (Araneae, Coleoptera, Diptera, Hemiptera, Hymenoptera, and Lepidoptera). We then assessed for each species whether the peak of emergence was identical in time (within the same sample period) between and within each farming type (OG˄IG, OF˄IF, OG˄OF, and IG˄IF). Counts of the numbers of these joint emergences in comparison with the total numbers of joint occurrences indicate similar or divergence (habitat specific) phenology.

| Biomass
Weight of wet and dry biomass significantly correlated across all samples ( Table 1, Table S1). Malaise traps set in the organic farmland

| Community composition with respect to OTUs
For all sufficiently rich taxa, except for Araneae and Orthoptera, we found significantly lower numbers of joint OTUs than what would be expected from a random sample model, assuming complete sampling of the local species pool (Table 5). However, even a small degree of undersampling (<10%; Table 5) was in accordance with the observed degree of joint species. These results were corroborated by the high and significant rank correlations between the jointly occurring species at both study sites indicating highly similar community composition and abundance distributions between the common farmland species (Table 5). The proportions of species occurring only in the organic farmland were higher than those in the conventional farmlands except of the species-poor taxa Araneae, Orthoptera, and Psocoptera ( Figure 2). One-way ANOVA showed that the absence of an OTU was highly significantly linked to its overall abundance. Rare OTUs were most prone to becoming extinct in one of the farms (F (1, 3,892) = 89.01, p < .001).

| Species communities
Altogether, 47 species of the Bavarian "Red Book of threatened Lepidoptera species" (Bayerisches Landesamt für Umweltschutz, LFU, 2003; were found in the Malaise traps and at light (Tables S3 and S4 The species-specific analysis showed that more than 50% (in most comparisons more than 65%) of species had identical peaks of emergence across the habitat types (Table 6). Both grasslands had less similar species emergence peaks than both forest fringes (Table 1). Importantly, organic grasslands and forest fringes were more similar in species phenology than the intensively managed grasslands and fringes (Table 6).

| Complementing quantitative and qualitative assessment
The combination of a quantitative and qualitative data set provides various advantages, while each single method shows different strengths, but also shortcomings. DNA metabarcoding allows analyses of large arthropod bulk samples of several thousand individuals from hundreds to thousands of species within only few weeks (Creedy, Ng, & Vogler, 2019;Morinière et al., 2016), including comprehensive species inventories (Wang, Srivathsan, Foo, Yamane, & Meier, 2018;Yu et al., 2012), species turnover (Doi et al., 2016), species composition in transects (Ji et al., 2013), seasonal fluctuations (Figure 2), and other aspects. This is a major advantage, especially when assessing insect species, which frequently occur in high densities and represent a very diverse group of organisms, with a large number of species difficult to identify, or which are still not yet described (Page, 2016;Stork, 2018). Traditional, morphology-based, "manual" identification of such large amounts of insect samples is time-and cost-consuming, and thus usually unrealistic as it is shown by a recent analysis of German malaise traps through experts requiring nine years for performing sorting and a morphology-based identification of roughly 10% of the collected individuals (Ssymank & Doczkal, 2017). Thus, an automated, time-, and cost-efficient system is the prerequisite to establish a large-scale longterm insect monitoring. data on the number of species and the amount of biomass provide little information on the status of an ecosystem or of entire landscapes, and information on the abundance of species and species compositions is crucial (Elbrecht & Leese, 2015). Thus, additionally we performed selective sampling of a well-known taxonomic group to assemble supplementary data to our DNA metabarcoding approach. Nocturnal lepidopterans are a suitable group to evaluate ecosystem health and the status of entire landscapes (Holloway, 1980). Nocturnal lepidopterans are very rich in species and comparatively well understood in terms of taxonomy and ecology (Haslberger & Segerer, 2016;Hering, 1951;Kristensen, 1999;Scoble, 1995). The use of further ecological characteristics for single species provides very relevant information on potential community shifts (e.g., reductions of evenness, decline of species with specific habitat requirements, shifts in species composition, Habel et al., 2016). Thus, this second, qualitative approach yields detailed information on species abundance and the structure and quality of a species community. In addition, with the use of light traps we obtain information from a larger, landscape scale, as this method attracts individuals from adjoining habitats, and thus, the collected samples rather reflect the surrounding area, than only from the grassland patch on which data collection was conducted (Truxa & Fiedler, 2012).

| Organic versus conventional farming
We sampled insects in organic and conventional farmland. Our data reveal higher biomass and species richness, and twice the number of threatened nocturnal lepidopterans in organic farmland if compared with our sampling sites in conventional farmland. However, we have to interpret these findings with caution, as data collection was performed on only two sites of each farming type. Thus, we only may conclude vague trends, but cannot (yet) derive very meaningful conclusions. However, our trends are in accordance with other studies showing a significant loss of biomass and reduced species richness in conventional farmland (Sanders & Hess, 2019). While we found significant differences of arthropod biomass and species richness between the two farmland types, we found similar species community composition across all four sampling sites, which is congruent with other studies (Gibson, Pearce, Morris, Symmondson, & Memmott, 2007), but also contrasts with previous work (Boutin, Martin, & Baril, 2009;Tsutsui, Kobayashi, & Miyashita, 2018). Our data show that Red List species mainly occur in organic farmland, which goes in line with Sanderson-Bellamy, Svensson, Brink, Gunnarsson, and Tedengren (2018), showing that specialist species (frequently also found on Red Lists) suffer in particular under agricultural intensification. TA B L E 5 Results of the random sample model for expected and observed numbers of OTUs that were found at both study sites Note: Also given are the total numbers of OTUs to obtain the observed number of joint occurrences at the 5% error level. Spearman's r provides the rank-order correlation in abundances of species jointly occurring between organic and conventional farming sites.

CO N FLI C T O F I NTE R E S T
TG, AG, and JK are permanent employees of the HIPP Company which operates the organic farm, one of the two study sites. The other authors declare no competing interests.

AUTH O R CO NTR I B UTI O N S
Thomas Greifenstein, Armin Günter, and Dieter Doczkal con- Collection of Zoology (ZSM) (Munich, Germany). All molecular data obtained from metabarcoding are provided in Table S2, as well as as interactive Krona files (Figures S1-S4). All raw data used in this article are uploaded to the public repository Dryad, accessibly under https://doi.org/10.5061/dryad.mpg4f 4qvw.